def add_prediction_op(self):
x = self.add_embedding(
) # [batch_size, num_time_steps, embedding_size]
cell = tf.nn.rnn_cell.BasicLSTMCell(self.config.rnn_hidden_size)
outputs, state = tf.nn.dynamic_rnn(cell,
x,
time_major=True,
dtype=tf.float32)
x = outputs[:, -1, :]
xavier_initializer = xavier_weight_init()
W = tf.Variable(
xavier_initializer(
[self.config.rnn_hidden_size, self.config.hidden_size_1]))
b1 = tf.Variable(
tf.zeros([self.config.hidden_size_1], dtype=tf.float32))
U = tf.Variable(
xavier_initializer(
[self.config.hidden_size_1, self.config.n_classes]))
b3 = tf.Variable(tf.zeros([self.config.n_classes], dtype=tf.float32))
layer_1 = tf.add(tf.matmul(x, W), b1)
h_drop = tf.nn.dropout(layer_1, self.dropout_placeholder)
pred = tf.matmul(h_drop, U) + b3
self.regularization = self.config.reg * tf.nn.l2_loss(
U) + self.config.reg * tf.nn.l2_loss(W)
return pred
def add_prediction_op(self):
x = self.add_embedding()
xavier_initializer = xavier_weight_init()
W = tf.Variable(
xavier_initializer([
self.config.n_window_features * self.config.embed_size,
self.config.hidden_size_1
]))
b1 = tf.Variable(
tf.zeros([self.config.hidden_size_1], dtype=tf.float32))
U = tf.Variable(
xavier_initializer(
[self.config.hidden_size_2, self.config.n_classes]))
W2 = tf.Variable(
xavier_initializer(
[self.config.hidden_size_1, self.config.hidden_size_2]))
b2 = tf.Variable(
tf.zeros([self.config.hidden_size_2], dtype=tf.float32))
b3 = tf.Variable(tf.zeros([self.config.n_classes], dtype=tf.float32))
layer_1 = tf.add(tf.matmul(x, W), b1)
layer_1 = tf.nn.sigmoid(layer_1)
layer_2 = tf.add(tf.matmul(layer_1, W2), b2)
layer_2 = tf.nn.sigmoid(layer_2)
h_drop = tf.nn.dropout(layer_2, self.dropout_placeholder)
pred = tf.matmul(h_drop, U) + b3
# self.regularization = self.config.lr*tf.nn.l2_loss(U) + self.config.lr*tf.nn.l2_loss(W)
return pred
def add_prediction_op(self):
x = self.add_embedding()
cell = tf.nn.rnn_cell.BasicLSTMCell(self.config.rnn_hidden_size)
outputs, state = tf.nn.dynamic_rnn(cell, x, time_major=True, dtype=tf.float32)
x = outputs[:,-1,:]
xavier_initializer = xavier_weight_init()
W1 = tf.Variable(xavier_initializer([self.config.rnn_hidden_size, self.config.hidden_size_1]))
b1 = tf.Variable(tf.zeros([self.config.hidden_size_1], dtype=tf.float32))
W2 = tf.Variable(xavier_initializer([self.config.hidden_size_1, self.config.hidden_size_2]))
b2 = tf.Variable(tf.zeros([self.config.hidden_size_2], dtype=tf.float32))
W3 = tf.Variable(xavier_initializer([self.config.hidden_size_2, self.config.n_classes]))
b3 = tf.Variable(tf.zeros([self.config.n_classes], dtype=tf.float32))
layer_1 = tf.add(tf.matmul(x, W1), b1)
layer_1 = tf.nn.relu(layer_1)
layer_2 = tf.add(tf.matmul(layer_1, W2), b2)
layer_2 = tf.nn.relu(layer_2)
distrib = tf.add(tf.matmul(layer_2, W3), b3) # not actually the probability distrib yet because using softmax_cross_entropy
pred = tf.nn.softmax(distrib)
self.regularization = self.config.reg*tf.nn.l2_loss(W1) + self.config.reg*tf.nn.l2_loss(W2) + self.config.reg*tf.nn.l2_loss(W3)
return pred
def add_prediction_op(self):
x = self.add_embedding()
xavier_initializer = xavier_weight_init()
W1 = tf.Variable(xavier_initializer([self.config.n_features*self.config.embed_size, self.config.hidden_size_1]))
b1 = tf.Variable(tf.zeros([self.config.hidden_size_1], dtype=tf.float32))
W2 = tf.Variable(xavier_initializer([self.config.hidden_size_1, self.config.hidden_size_2]))
b2 = tf.Variable(tf.zeros([self.config.hidden_size_2], dtype=tf.float32))
W3 = tf.Variable(xavier_initializer([self.config.hidden_size_2, self.config.n_classes]))
b3 = tf.Variable(tf.zeros([self.config.n_classes], dtype=tf.float32))
layer_1 = tf.add(tf.matmul(x, W1), b1)
layer_1 = tf.nn.relu(layer_1)
layer_2 = tf.add(tf.matmul(layer_1, W2), b2)
layer_2 = tf.nn.relu(layer_2)
distrib = tf.add(tf.matmul(layer_2, W3), b3) # not actually the probability distrib yet because using softmax_cross_entropy
pred = tf.nn.softmax(distrib)
self.regularization = self.config.reg*tf.nn.l2_loss(W1) + self.config.reg*tf.nn.l2_loss(W2) + self.config.reg*tf.nn.l2_loss(W3)
return pred